The neural computation technology is often unsed for chemical pattern classification. In is rather difficult to apply neural networks for classifying complex chemical pattern, which has the property of high-dimension but low-sample- number. By extracting pattern characteristic, decreasing the dimension of network input, this problem in complex pattern classification can be relatively easily solved. Based on the principal of searching class correlative component a new method, named stepwise class correlative components analysis (SCCCA), is proposed. The technique can extract characteristic component that has relatively large correlative value with the class measurement from the orginal dataset. Comparing with principal component analysis (PCA), a typical example in identifying the composition-activity relationship of a natural blant was used, and the results verified that the new method is better than PCA.

Key words: NATURAL PRODUCTS, NEURONS, NEURAL NETWORK